Attention-based adaptive intra refresh for error-prone video transmission

摘要

Error-resilient video coding plays an important role in video transmission over error-prone networks. Intra coding, which can suppress error propagation very well in a simple way, is often employed in error-resilient video coding in some special ways and named as intra refresh. However, most of the existing intra refresh methods do not make good use of subjective human vision property. In this article a novel intra refresh method is presented, taking the fact that people usually intend to pay much more attention to a certain particular area in a video frame (attention area) than to other areas (nonattention area) in the same frame. In this method, blocks in the attention area of each video frame have a higher priority to be refreshed as intra mode than the blocks outside the attention area. Meanwhile, an attention-based end-to-end rate distortion model with reference restriction for both inter coding and intra coding is developed to obtain a better subjective quality. Experimental results in H.264/AVC-based video coding demonstrate that, when compared at the same bit rate and the same packet loss condition, the proposed method can provide a much better subjective feeling than existing intra refresh methods